Learning and Simulating the Earth’s Water Cycle with NCCS Resources

Leveraging NASA Center for Climate Simulation (NCCS) resources, University of Alabama and NASA Goddard Space Flight Center researchers are integrating machine learning techniques into the NASA Land Information System (LIS) data assimilation framework to help reduce uncertainties in the simulation model and to account for uncertainties in the data.

They use machine learning to improve LIS’s representation of key ecosystem feedback processes by learning from differences between satellite observations and model predictions and ultimately better predict the global water cycle and monitor water’s role in Earth’s ecosystems.

Last year, the researchers used approximately 6.5 million core-hours on the NCCS Discover supercomputer. To reduce the cost of integrating machine learning into the simulation runs, they parallelized Gaussian Process Regression training routines that resulted in advantageous scaling up to thousands of processor cores. The simulations also used Discover online storage.

A video comparing three types of machine learning and linear regression for blending several remote sensing data products to estimate monthly terrestrial ecosystem carbon fluxes globally. Research by Grey Nearing,Jonathan Frame, Donovan Murphy, University of Alabama; Craig Pelissier, NASA Goddard Space Flight Center; and Milton Halem, University of Maryland, Baltimore County.

So far, this project has created new data fusion estimates of global terrestrial carbon uptake from natural ecosystems, which are being used to improve model calibrations for flood forecasting across the continental U.S.

“Hydrology models must generally be run over large spatial domains, and assimilating satellite data requires running large ensembles,” said Grey Nearing assistant professor of geological sciences at the University of Alabama. These models are already computationally expensive, and the expense increases when we add the training and optimization routines necessary to learn new patterns and connections in large remote-sensing data records. Our work would be impossible without the large computational resources at the NCCS.”

More information

NASA@SC18: Learning and Simulating the Earth’s Water Cycle